OpenNMT-py pause training without errors and no stopping


I observed that training gets paused without continuing and no error information comes.
I am using Transformer model and early stopping was in patience 7/10. I started with train_from a previous model.


This is presumably fixed with the latest PyTorch version. Try OpenNMT-py 0.9.2 with PyTorch 1.2.

I using pytorch 1.6.0+cu101 and OpenNMT-py 1.2.0(latest version)
I found that i have problem like @merc85garcia’s issue last week
For small data size;( size is 34M), I made models successfully but for bigger train data size, i didn’t that

Would I select method that @guillaumekln recommended to merc85gracia ?
If you know my train configure and others, i’ll tell you

+) When I change configure 4gpu to only one gpu, it’s worked without pausing
probably I think that this issue is involved with gpu memory or gpu exclusive mode?